icount: improve exec nocache usage

cpu-exec tries to execute TB without caching when current
icount budget is over. But sometimes refilled budget is big
enough to try executing cached blocks.
This patch checks that instruction budget is big enough
for next block execution instead of just running cpu_exec_nocache.
It halves the number of calls of cpu_exec_nocache function
during tested OS boot scenario.

Signed-off-by: Pavel Dovgalyuk <pavel.dovgalyuk@ispras.ru>
---
 accel/tcg/cpu-exec.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

On 08/12/20 10:10, Pavel Dovgalyuk wrote:
> cpu-exec tries to execute TB without caching when current
> icount budget is over. But sometimes refilled budget is big
> enough to try executing cached blocks.
> This patch checks that instruction budget is big enough
> for next block execution instead of just running cpu_exec_nocache.
> It halves the number of calls of cpu_exec_nocache function
> during tested OS boot scenario.
> 
> Signed-off-by: Pavel Dovgalyuk <pavel.dovgalyuk@ispras.ru>
> ---
>   accel/tcg/cpu-exec.c |    2 +-
>   1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/accel/tcg/cpu-exec.c b/accel/tcg/cpu-exec.c
> index 58aea605d8..251b340fb9 100644
> --- a/accel/tcg/cpu-exec.c
> +++ b/accel/tcg/cpu-exec.c
> @@ -685,7 +685,7 @@ static inline void cpu_loop_exec_tb(CPUState *cpu, TranslationBlock *tb,
>       insns_left = MIN(0xffff, cpu->icount_budget);
>       cpu_neg(cpu)->icount_decr.u16.low = insns_left;
>       cpu->icount_extra = cpu->icount_budget - insns_left;
> -    if (!cpu->icount_extra) {
> +    if (!cpu->icount_extra && insns_left < tb->icount) {
>           /* Execute any remaining instructions, then let the main loop
>            * handle the next event.
>            */
> 

Queued, thanks.

Paolo

On 12/8/20 3:10 AM, Pavel Dovgalyuk wrote:
> cpu-exec tries to execute TB without caching when current
> icount budget is over. But sometimes refilled budget is big
> enough to try executing cached blocks.
> This patch checks that instruction budget is big enough
> for next block execution instead of just running cpu_exec_nocache.
> It halves the number of calls of cpu_exec_nocache function
> during tested OS boot scenario.
> 
> Signed-off-by: Pavel Dovgalyuk <pavel.dovgalyuk@ispras.ru>
> ---
>  accel/tcg/cpu-exec.c |    2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/accel/tcg/cpu-exec.c b/accel/tcg/cpu-exec.c
> index 58aea605d8..251b340fb9 100644
> --- a/accel/tcg/cpu-exec.c
> +++ b/accel/tcg/cpu-exec.c
> @@ -685,7 +685,7 @@ static inline void cpu_loop_exec_tb(CPUState *cpu, TranslationBlock *tb,
>      insns_left = MIN(0xffff, cpu->icount_budget);
>      cpu_neg(cpu)->icount_decr.u16.low = insns_left;
>      cpu->icount_extra = cpu->icount_budget - insns_left;
> -    if (!cpu->icount_extra) {
> +    if (!cpu->icount_extra && insns_left < tb->icount) {

Reviewed-by: Richard Henderson <richard.henderson@linaro.org>

I also wonder if we should really be not caching these.  Ever since MTTCG, we
have not actually been reusing the memory.  We're simply removing the TB from
the hash table.  I think we should be remembering these just in case we can in
fact reuse them.


r~

On 12.12.2020 00:41, Richard Henderson wrote:
> On 12/8/20 3:10 AM, Pavel Dovgalyuk wrote:
>> cpu-exec tries to execute TB without caching when current
>> icount budget is over. But sometimes refilled budget is big
>> enough to try executing cached blocks.
>> This patch checks that instruction budget is big enough
>> for next block execution instead of just running cpu_exec_nocache.
>> It halves the number of calls of cpu_exec_nocache function
>> during tested OS boot scenario.
>>
>> Signed-off-by: Pavel Dovgalyuk <pavel.dovgalyuk@ispras.ru>
>> ---
>>   accel/tcg/cpu-exec.c |    2 +-
>>   1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/accel/tcg/cpu-exec.c b/accel/tcg/cpu-exec.c
>> index 58aea605d8..251b340fb9 100644
>> --- a/accel/tcg/cpu-exec.c
>> +++ b/accel/tcg/cpu-exec.c
>> @@ -685,7 +685,7 @@ static inline void cpu_loop_exec_tb(CPUState *cpu, TranslationBlock *tb,
>>       insns_left = MIN(0xffff, cpu->icount_budget);
>>       cpu_neg(cpu)->icount_decr.u16.low = insns_left;
>>       cpu->icount_extra = cpu->icount_budget - insns_left;
>> -    if (!cpu->icount_extra) {
>> +    if (!cpu->icount_extra && insns_left < tb->icount) {
> 
> Reviewed-by: Richard Henderson <richard.henderson@linaro.org>

Thanks.

> 
> I also wonder if we should really be not caching these.  Ever since MTTCG, we
> have not actually been reusing the memory.  We're simply removing the TB from
> the hash table.  I think we should be remembering these just in case we can in
> fact reuse them.

I'm still thinking about reusing these blocks. Sometimes there are 
loops, where blocks of small sizes like 1..3 are translated for many times.

However, we can't cache them directly, because hash table can include 
only one block with the specific pc.

Pavel Dovgalyuk

On 12/12/20 12:22 AM, Pavel Dovgalyuk wrote:
> However, we can't cache them directly, because hash table can include only one
> block with the specific pc.

That's not true at all.


r~

On 12.12.2020 17:31, Richard Henderson wrote:
> On 12/12/20 12:22 AM, Pavel Dovgalyuk wrote:
>> However, we can't cache them directly, because hash table can include only one
>> block with the specific pc.
> 
> That's not true at all.

Thanks, I missed, that icount is used for TB comparison in hash container.

Pavel Dovgalyuk